Thanks for considering a contribution! This skill is evidence-driven — every claim about a model's behavior should be traceable to either an official doc, a community post (link the source), or a reproducible test.
- New
automation/site-profiles/<platform>.md— for platforms currently marked ⏳ (unverified) - Model-version updates — when a platform ships a new version, the signature token list often shifts
- Better presets in
templates/preset-packs.md - Counter-examples — if a "rule" in this skill produces bad output on your task, file an issue with the prompt + screenshot
This is the cross-platform research consolidation. When adding a model, include:
- Structure formula (
Subject + Action + ...) - Length sweet spot (token / character count)
- Signature tokens (positive list)
- Taboos / known-bad patterns (negative list)
- Source links (3+ independent sources preferred)
- Markdown. No fancy HTML.
- Code blocks for prompts and CLI commands.
- Tables for per-model comparisons.
- Cite sources inline where the claim originated.
- Trad. Chinese OR English in body text is fine — but prompt examples should match the platform's native language preference (Seedance / Wan = Chinese; MJ / Flux / Sora 2 = English).
- Prompts that copy copyrighted lyrics / scripts
- Prompts targeting deepfakes of real people without context
- "Workaround" instructions for bypassing a platform's safety filters
- Personally identifiable information from any platform (account names, workspace UUIDs, payment info)